Multi-Region Active-Active Architecture

Even with availability zones, a workload deployed in a single Azure region can be affected by a regional outage — a rare but possible event caused by a major disaster, severe storm, or large-scale infrastructure failure. For the highest levels of availability, enterprises deploy workloads across two or more Azure regions, allowing the application to continue serving users even if an entire region goes offline.

There are two main multi-region patterns:

• Active-Active — resources in both regions serve live traffic simultaneously. This maximises availability and can also reduce latency by routing users to the nearest region.
• Active-Passive — resources in the secondary region are on standby and only receive traffic after a failover event. This costs less but has a longer recovery time.

For the highest availability, active-active is the preferred approach.

Azure Traffic Manager is a DNS-based global load balancer that distributes traffic across endpoints in different Azure regions (or on-premises). It uses health probes to monitor each endpoint and routing methods (performance, weighted, priority, geographic, multi-value) to decide where to send traffic. For an active-active setup, use the performance routing method to send users to the nearest healthy region.

Traffic Manager supports several routing methods tailored to different scenarios:

• Performance — routes to the lowest-latency endpoint (best for active-active)
• Weighted — distributes traffic by percentage weight (useful for gradual rollouts)
• Priority — sends all traffic to the primary endpoint, with fallback to secondary (active-passive)
• Geographic — routes users based on their geographic location (data sovereignty)

In an active-active architecture, data must be replicated between regions so that users in either region read consistent data. Options include:

• Azure SQL Database with geo-replication — readable secondary in another region
• Azure Cosmos DB with multi-region writes — all regions accept writes simultaneously
• Azure Storage with GRS — async replication to a secondary region

The choice depends on your consistency and write-availability requirements.

Azure Cosmos DB is uniquely suited for active-active architectures because it supports multi-region writes — all configured regions can accept write operations simultaneously. Cosmos DB uses configurable consistency levels (from strong to eventual) to manage how replicas are synchronised. For active-active scenarios, session consistency or bounded staleness are common choices that balance performance with correctness.

Both Traffic Manager and Azure Front Door provide global load balancing, but at different levels:

• Traffic Manager is DNS-based — it redirects the client to the correct regional endpoint, but the connection goes directly to that endpoint. It operates at the DNS layer.
• Azure Front Door is an anycast proxy — it terminates the client connection at a global PoP and forwards the request to the origin. It provides WAF, SSL offload, and caching in addition to routing.

Microsoft designates region pairs for each Azure region — a secondary region within the same geography that is geographically distant enough to survive regional disasters. When designing multi-region architectures, use the paired region as your secondary where possible. During planned platform maintenance, only one region in a pair is updated at a time, reducing simultaneous disruption.

In active-active deployments, a user's request may be served by different regions on different requests. If your application stores session state in memory on the web server, this causes problems when a user switches regions. The solution is to use a distributed session store — such as Azure Cache for Redis with geo-replication — so session data is accessible from any region.

An active-active architecture is more expensive than a single-region deployment because you are running a full copy of your infrastructure in at least two regions. Additional costs include inter-region data transfer, geo-replication for databases, and Traffic Manager or Front Door charges. Conduct a cost-benefit analysis: for mission-critical workloads, the cost of downtime typically far exceeds the cost of the second region.

An active-active architecture must be tested regularly to ensure it works when needed. Chaos engineering tools like Azure Chaos Studio allow you to inject failures — such as stopping all VMs in one region — and observe how the system responds. Regular testing confirms that Traffic Manager health probes detect failures, failover happens within your RTO, and data remains consistent across regions.

Test your understanding of Microsoft Azure Fundamentals (AZ-900) concepts from this lesson.

In this lesson you learned: active-active architecture deploys live resources in multiple regions simultaneously for maximum availability; Azure Traffic Manager routes traffic using performance, priority, weighted, or geographic methods; and data replication (Cosmos DB multi-write, SQL geo-replication) is essential for consistency across regions. Next up we explore health probes and graceful degradation patterns.